Citigroup rolls out AI agent platform Arc in Hong Kong amid US-China tensions

Share:
Audio Loading voice…
Citigroup rolls out AI agent platform Arc in Hong Kong amid US-China tensions

Synopsis

Citigroup has quietly launched an AI agent platform called Arc in Hong Kong even as OpenAI and Anthropic block model access to the city — revealing how Wall Street is building proprietary AI infrastructure to sidestep geopolitical restrictions without slowing adoption.

Key Takeaways

Citigroup began trialling its in-house AI agent platform Arc in Hong Kong in May 2026 , initially targeting software developers and engineers.
Arc automates complex tasks such as client prospecting, which previously required hours of manual data gathering and scenario modelling.
Senior management at Citigroup Hong Kong has been 'aggressively' encouraging AI adoption since summer 2025 , according to an anonymous employee.
OpenAI and Anthropic have both withheld model access for users in Hong Kong and mainland China , complicating bank-wide AI rollouts.
Peers including Goldman Sachs and Morgan Stanley are also advancing AI deployments in the city, with tools including Google Gemini -based platforms under evaluation.

Citigroup has launched an in-house AI agent platform called Arc at its Hong Kong offices this month, becoming one of the clearest examples of American banks pushing ahead with US-powered artificial intelligence tools in the city despite intensifying geopolitical friction. The rollout initially targets local software developers and engineers, according to a source familiar with the matter.

What Arc does and why it matters

Arc represents what Citigroup describes as a step-change improvement over its previous in-house AI tools, enabling greater end-to-end automation of complex workflows. Tasks such as client prospecting — which typically require hours of gathering portfolio data, analysing market trends, and modelling scenarios — can now reportedly be handled with far less manual intervention.

An anonymous Citigroup employee in Hong Kong said senior management had been 'aggressively' encouraging AI adoption since last summer. 'AI agents are the new thing that really makes me feel like everything could change,' the person said. 'Previously, I would have to prompt the AI through multiple turns, but now it looks like the AI agents can do this on their own.'

The compliance minefield

The aggressive adoption comes against a fraught regulatory and geopolitical backdrop. US AI giants OpenAI and Anthropic — maker of the Claude model family — have both withheld access to their models for users based in Hong Kong and mainland China, forcing financial institutions to navigate a complex compliance landscape.

That access gap means banks cannot simply deploy off-the-shelf frontier models and must instead build or license proprietary systems, raising questions about model parity and long-term competitiveness relative to peers operating in unrestricted markets.

Broader Wall Street push in Hong Kong

Citigroup is not alone. Peers including Goldman Sachs and Morgan Stanley are also reported to be advancing AI deployments in the city, while platforms such as CyberMind and tools built on Google Gemini are being evaluated across the sector. The Hong Kong Monetary Authority has been engaging with financial institutions on responsible AI adoption frameworks, adding a layer of regulatory oversight to the rollout race.

The moves collectively signal that Wall Street views Hong Kong as a sufficiently strategic market to justify the operational complexity of building access-restricted AI infrastructure.

What's next

The immediate question is whether Arc and similar proprietary platforms can match the capability trajectory of frontier models that remain off-limits in the territory. As US-China tensions evolve, any further tightening of technology export controls or model-access restrictions could force banks to accelerate investment in alternative AI stacks. Watch for signals from the Hong Kong Monetary Authority on formal AI governance guidelines, which could shape how aggressively institutions like Citigroup scale these deployments citywide.

Point of View

Creating a two-tier capability landscape. What mainstream coverage tends to underplay is that this is not merely a compliance headache — it is an accelerant for in-house AI engineering capacity at major financial institutions, which may ultimately reduce their dependence on OpenAI and Anthropic regardless of how access restrictions evolve. The broader pattern mirrors what happened in semiconductor supply chains: restrictions intended to contain technology diffusion end up spurring parallel development. If proprietary bank AI agents reach competitive parity with restricted frontier models, the leverage that US AI firms currently hold over financial-sector clients in restricted markets shrinks considerably.
NationPress
13 Jul 2026

Frequently Asked Questions

What is Citigroup's Arc AI platform in Hong Kong?
Arc is an in-house AI agent platform launched by Citigroup in Hong Kong in May 2026 . It automates complex, multi-step tasks such as client prospecting by handling data gathering, market analysis, and scenario modelling with minimal human prompting, representing a significant upgrade over the bank's previous internal AI tools.
Why can't Hong Kong users access OpenAI and Anthropic models?
OpenAI and Anthropic have both chosen not to make their AI models available to users based in Hong Kong and mainland China , reportedly due to geopolitical considerations and compliance concerns. This forces banks and other enterprises in the city to rely on proprietary or alternative AI systems.
Which other US banks are deploying AI in Hong Kong?
Goldman Sachs and Morgan Stanley are among the US banks also advancing AI deployments in Hong Kong , alongside Citigroup . Platforms including CyberMind and tools built on Google Gemini are reportedly being evaluated across the sector.
What role is the Hong Kong Monetary Authority playing in bank AI adoption?
The Hong Kong Monetary Authority has been engaging with financial institutions on responsible AI adoption frameworks. Its formal guidelines are expected to shape how aggressively banks can scale AI agent deployments across their Hong Kong operations.
How do AI agents differ from standard AI chatbots for banking tasks?
Unlike standard AI tools that require users to issue multiple sequential prompts, AI agents can autonomously execute multi-step workflows end-to-end. A Citigroup employee described the shift: 'Previously, I would have to prompt the AI through multiple turns, but now it looks like the AI agents can do this on their own.'
Nation Press
The Trail

Connected Dots

Tracing the thread behind this story — newest first.

8 Dots
  1. Latest 1 week ago
  2. 3 weeks ago
  3. 4 weeks ago
  4. 1 month ago
  5. 1 month ago
  6. 1 month ago
  7. 1 month ago
  8. 1 year ago
Google Prefer NP
On Google